RTX 2080 TI GPU Acceleration for Deep Learning Toolbox?

Hi,

We have one laptob and one workstation in our lab. Laptop has nvidia quadro p600 and workstation has nvidia rtx2080 ti gpu on it. I know that Matlab 2018b deep learning toolbox implements single precision operations for GPU by default. Processing power is 11 750.40 GFLOPS for RTX2080 ti and 1195 GFLOPS for Quadro P600 (defined in here) for single precision. But training time of quadro p600 of is 4x faster than rtx2080 on deep learning tooolbox for the same program. I think that rtx2080ti shold be much and much faster than quadro p600. I'm really confused. How could I accelerate training time for RTX2080 ti? Which Matlab settings must I change for this?

Would you give me advices please?

Best Regards...

[Ive J]

This is interesting (I also have RTX 2080 ti!). Have you checked with profile as well to see maybe something else is going on there? Also make sure that MATLAB recognizes your GPU (gpuDevice), and if so, set ExecutionEnvironment of trainingOptions to gpu.

[ahmet emir]

Firstly, thanks for your answer , yes I tried gpuDevice command. MATLAB recognizes my GPU. Yes I set ExecutionEnvironment of trainingOptions to gpu. But why is it slow I couldn't unterstand. I will checked training with profiler according to your advice

[ahmet emir]

Which parameters should I control in profiler viewer? Would you give me advices??

[Ive J]

Oops, I misspelled it, it's profile. Put it around your code block like this:

profile on

% your training snippet

profile viewer

[ahmet emir]

yes I had understood. . I have used above code. But I'm not decided that which results should I change to accelerate RTX 2080 Tİ. Same codes I used both Quadro p600 and RTX2080 Tİ but unfortunately p600's training time is lower than training time of RTX2080 TI.

[Ive J]

In your question you worte P600 was faster, but now you say the opposite. profile of the same script (as yours) should be fairly the same between two systems (assuming other configs are similar).

[ahmet emir]

sorry you are correct p600 is much faster

[ahmet emir]

I want to try MATLAB GPU benchmark program ( here) . Laptop with quadro p600 gpu runned this program without error and the gpu testing process was completed. It works for laptop. But Z820 restart on every run of gpu bencmark after %11 progressing of testing. I'm really confused. How I come of this problem?? I set up Cuda 9.1 toolkit , RTX2080TI drivers correctly.